POTENTIAL APPLICATIONS

Regional thresholds (e.g., Atlanta) are used to account for the driving context and highlight

extreme driving. Two types of driving volatility information can be provided to drivers:

 Real time driving behavior information: Drivers may be alerted or warned when they exceed certain thresholds of acceleration or vehicular jerk, providing them with dynamic

feedback on their volatility through Advanced Traveler Information Systems (ATIS).

Displays can be designed to inform drivers their real-time driving volatility, without

overly distracting them, e.g., through a light on the dashboard that turns yellow or red

from green. This can also be supplemented via email notifications.

 Daily/monthly/yearly driving behavior summary information. Long-term advice on driving patterns can be provided to the driver based on analysis of their daily, monthly or

yearly driving performance. Such information can be provided through websites, and

may contain a record, analysis of driving patterns and customized advice on improving

accelerations, braking, speeds, and turns, etc.

Thresholds of identifying extreme driving patterns can be based on combinations of

accelerations, single vehicular jerk, expanded vehicular jerk and variance in these parameters

[51]. While this study used the mean plus/minus one standard deviation thresholds for

identifying extreme patterns, other threshold criteria can also be used, e.g., mean plus two or

three standard deviations. Note that, the thresholds may be further adjusted based on time of day,

weather, terrain, and roadway classification. They can be personalized based only on trips

undertaken by the individual or use regional data to calculate thresholds. Adding these functions

3.8 LIMITATIONS

This study depends heavily on GPS data collected by in-vehicle devices. To some extent the

accuracy and availability of location data constrain the analysis. Compared with high industrial

sampling rates (e.g. 96 kHz), these data are limited by relatively low sampling frequency which

gives only second-by-second speeds. A reasonable question is whether second-by-second speed

data are good enough for identifying instantaneous driving decisions. To address this issue,

additional analyses were conducted by collecting driving data at 20 Hz using a driving simulator

[83]. This database includes 35,924 seconds speed data made by 24 drivers, generating 718,481

speed data points, which allows the investigation of micro-driving decision changes within one

second. The results show that drivers made no change to their speed for 89.9% of the sampled

seconds, i.e., drivers either kept accelerating, decelerating or just maintained speed during a

second. Only 10.1% of the sampled seconds involve driver’s decision change. Overall, the

analysis found that at least 98.5% instantaneous driving decision changes can be detected using

second-by-second data compared with smaller intervals and that the second-by-second data are

reasonably accurate for the purposes of this study.

Some other critical information remains unknown to the researchers due to privacy concerns.

This includes the type of roads and the geo-codes for each second of driving. Missing

geographically referenced information for trips prevents the researchers from extracting useful

contextual factors. These include roadway segments used during trips and associated traffic

counts, road geometry, traffic operations facilities, and surrounding land uses. Therefore, how

the instantaneous decisions are associated with surrounding traffic, facility and land use can be

volatility in instantaneous driving decisions. More research is needed to investigate the impacts

of network attributes, environmental attributes on instantaneous decisions, as shown in the

conceptual framework. Expansion of the study can form the basis of future analysis of driver

volatility and how it relates to energy, environment and safety.

3.9 CONCLUSIONS

In the context of using large-scale data for traffic safety improvement, tailpipe emissions and

energy use reduction in a driving dominant environment, it is essential to understand drivers’

instantaneous driving decisions and their associated impacts. The research takes advantage of

large-scale driving databases coupled by second-by-second GPS data to develop a framework for

the research agenda in driving behavior studies addressing how to define the instantaneous

driving decisions in a quantifiable way and how to quantify explicitly volatile driving in a

defensible manner. The answer is to create a volatility indicator to measure the gap between an

individual’s driving practice and the typical driving practice in that region. Assuming the typical driving practice applied by most people represents the norm of driving culture in that region, the

driving practices standing out of that norm could be defined as volatile driving. The paper

demonstrates a methodology to measure the volatility, which is based on variance in vehicular

jerk between individual drivers and regional sample profiles. The creation of a robust volatility

score that is able to quantify the extent of volatility, instead of simply labeling a driver as

aggressive or non-aggressive is a key contribution.

To create a typical driving profile for the study metropolitan area, acceleration or vehicular jerk

plus/minus one standard deviation). While typical driving practices are identified when the

acceleration or vehicular jerk fall between the bands, volatile driving is defined as accelerations

or vehicular jerks that fall out of the bands range. A volatility score for each trip or each driver

can be calculated by the percent of travel time spent on volatile driving. In this sense, developing

a regional driving profile is critical since this driving profile serves as a “standard” to define individual’s driving volatility. Atlanta’s driving profile was developed through an innovative visualization of data, the time spent on each driving behavior was calculated. Specifically,

overall 14% of the travel time spent on high vehicular jerk; 7% of driving time was spent on

idling or traveling at speeds below 5 mph, 47% of driving time was spent on acceleration, 41%

of driving time was spent on deceleration and 5% of driving time was spent on maintaining

constant speed. This information can be useful for designing driving cycle in a local context for

better emissions estimations. The methodology has great potential to be expanded to measure

driving volatility on road infrastructures as an indicator of roadway safety. Roads with higher

risk (those experiencing more hard braking and negative jerks) can be identified and proactive

strategies can be designed.

The findings are useful for potential applications to fleet vehicles and the general driving

population. Driving volatility information based on accelerations and vehicular jerk can be

incorporated in driving assist systems, e.g., advanced traveler information systems (ATIS).

Current traveler information systems (such as 511) are largely meant to support more macro

driver decisions (e.g., route choice and route diversion) and do not provide much instantaneous

information that can help drivers make more micro driving decisions. The real-time driving

vehicles or neighbors or just their own performance can support short-term micro decisions.

This in turn can benefit the community or fleets in several ways: 1) calmer driving; 2) safer

driving in general (especially on icy or slippery road surfaces where alert thresholds can be

lowered); 3) lower fuel consumption and emissions; and 4) identification of dangerous road